A framework predicting removal efficacy of antibiotic resistance genes during disinfection processes with machine learning.

Disinfection has been applied widely for the removal of antibiotic resistance genes (ARGs) to curb the spread of antibiotic resistance. Quantitative polymerase chain reaction (qPCR) is the most used method to quantify the damage of DNA thus calculating the ARG degradation during disinfection but suffers the deviation due to the limitation of amplicon length. In contrast, transformation assay more accurately measures ARG deactivation based on expression of disinfected ARG in the receiving bacteria but is typically laborious and material-intensive. This work applied machine learning (ML) to develop a framework by using qPCR results as a proxy to estimate the transformation assay measurements during disinfection with chlorine (FAC), ultraviolet (UV254), ozone (O3), and hydrogen peroxide/ultraviolet (UV/H2O2) for multiple kinds of ARGs. ARG degradation rates and deactivation rates were well predicted with the optimal correlation coefficient (R2) of all test sets > 0.926 and > 0.871, respectively. Besides, by concatenating the ARG degradation and deactivation predictive models, ARG removal efficiency under given disinfection conditions was directly predicted as the loss of transformation activity with R2 > 0.828. Furthermore, an online platform was built to provide users with access to the developed ML models for rapid and accurate evaluation of ARG removal efficiency.